Ensure FiberExit is thrown when scheduler destroyed

Different cleanup for uncaught exceptions

Enforce schedulers running only once

Use HAVE_VALGRIND to toggle valgrind support

Refactor some error handling

Using FiberExit since it’s uncatchable instead of a couple fatal errors, makes things a little nicer.

Fix fatal error handling

Fix use-after-free in uncaught exception handler

Add test with throw in finally

The behavior in this test matches that of a destroyed generator.

Handle potential ZEND_CATCH opcode handler from another extension

Properly declare globals for thread safety

Make an uncatchable exception by replacing catch opcode

This executes finally blocks as expected and is very simple. \o/

Revert to throwing UnwindExit

Left the code and tests handling finally blocks, we can revisit this later.

Throw from innermost try/catch

Vary cleanup opline

If execution is before catch, move execution to the catch block.

Remove misleading params

These functions work with the current EG, so the params were unnecesssary and misleading.

Locally backup EG data

Create zval in Fiber::run() for GC

This is required to properly trigger the GC to free the fiber memory.

Add failing test

Added test output files to .gitignore.

Execute finally blocks in unresumed fibers

Do not create zend_object for schedulers

The object is unneeded and has been giving me GC headaches from the start. This is so much simplier since I know when to GC the scheduler fiber.

GC root fiber

Fix GC

efree isn’t correct, but I’m not really sure what is yet…

Remove circular reference to GC unresumed fibers

Backup jit_trace_num

Only destroy on cleanup